メニュー
Surface concrete options to refine or refactor a given package. Read-only analysis; does not modify files.
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
SOC 職業分類に基づく
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/telorun/telo --skill refineコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
SKILL.md を表示中
Collaboratively plan a new feature. Gather context, ask questions, analyze together, then write a short high-level plan to ./plans only after explicit user approval.
Read-only code review for staged changes or current branch.
Read-only review of an implementation plan.
| name | refine |
| description | Surface concrete options to refine or refactor a given package. Read-only analysis; does not modify files. |
Do NOT modify any files. This is a read-only analysis task.
The user will supply a package path (or name) as $ARGUMENTS. If $ARGUMENTS is empty or ambiguous (e.g. just a name that matches multiple locations), ask the user to clarify before proceeding.
Resolve $ARGUMENTS to a concrete directory. Packages typically live under modules/<name>/, kernel/, cli/, sdk/, analyzer/, apps/, or ide/. If the argument is a bare name, check modules/<name>/ first. Confirm the resolved path before spawning the agent.
Use the Agent tool to spawn the refiner agent with the following prompt:
Examine the package at
<resolved path>and surface concrete options to refine or refactor it.Rules:
- Do NOT modify any files.
- Ground every option in specific file paths and line ranges from the package.
- Consider both Telo's core architectural goals (polyglot, visual editing, performance, static analysis) and general code-quality refinement categories (cohesion, duplication, abstraction fit, dead code, naming, error handling, test coverage, docs drift).
- Rate each option's impact and effort independently, sort by impact-to-effort ratio (best first), and call out quick wins (large impact, small effort) up top.
Read CLAUDE.md for architectural context.
Follow the response structure defined in your agent instructions (Package summary / Refinement options / Out of scope / Open questions).
Relay the refiner's output to the user verbatim, preserving its structure (keep the Quick wins section prominent). Do not collapse options and do not start implementing. After presenting, recommend starting with the quick wins if any were identified, then ask the user which option(s) they want to pursue — and wait for an explicit go-ahead before making any changes.